How Adrastia computes pharmacy reimbursement benchmark variance

Adrastia matches Medicaid and Medicare Part D reimbursement records to NADAC acquisition-cost benchmarks at the 11-digit NDC level, computes the percentage difference between the reimbursed unit price and the benchmark unit price, and ranks the results by magnitude. The output is a list of review candidates — records warranting further validation by the agency or payer — not a finding of wrongdoing.

Published 28 July 2026 · Last updated 30 July 2026

Definitions

Benchmark variance

The difference between the unit price a payer reimbursed and the published national average acquisition cost for that same drug product, expressed as a percentage of the benchmark.

Review candidate

A record whose variance is large enough to warrant validation against the payer's own claims data. A review candidate is a starting point for an agency-led review. It is not an allegation, a determination, or a recovery amount.

Unit basis

The denominator a price is expressed in — per tablet, per millilitre, per gram. Two prices are only comparable when both are stated on the same unit basis. Mismatched unit bases are the most common source of false variance in this kind of analysis, which is why unit normalization is a distinct validation step below.

Data sources and vintages

Every source below is a public federal file. Any figure Adrastia publishes carries the vintage of the files it was computed from, because the same query run against a different quarter returns a different number.

State coverage. The current build covers 48 SDUD-reporting states. Alaska, Vermont and the District of Columbia are not included. Adrastia does not describe its coverage as national while that remains true.

Step 1 — NDC matching and validation

Reimbursement records and benchmark records are joined on the 11-digit National Drug Code in 5-4-2 format. Before the join, codes are normalized: hyphen variants removed, segments zero-padded to 5-4-2, and 10-digit codes expanded to their 11-digit form.

Each matched code is then cross-validated against the FDA NDC Directory for active registration and therapeutic class. Where a code appears more than once, the retention rule is most-recent-quarter, highest-variance.

Step 2 — Variance computation

Records at or beyond ±100% are flagged as materially significant. Results are ranked by absolute magnitude, so both directions surface — a reimbursement materially below benchmark is as much a data-quality signal as one materially above it.

Worked example

The following arithmetic is illustrative. It uses round numbers to show the calculation and is not drawn from any state's data.

A record like this is a review candidate. It says the reimbursed price sits materially above the national average acquisition cost for that product in that period. It does not say the price was improper — dispensing fees, contractual arrangements, supply conditions, and product-specific factors all live outside these files, and the agency's own claims data is what resolves them.

Dispensing fees

An acquisition-cost benchmark does not include the pharmacy's dispensing fee. Where Adrastia reports net figures, a dispensing-fee allowance is applied and gross and net are shown side by side rather than one replacing the other. Any allowance applied in a client engagement is stated explicitly, and the intended source for it is the agency's own contracted fee schedule rather than a national average — dispensing fees are set by state and by contract, so a published figure from elsewhere is a placeholder, not a benchmark.

Step 3 — Sample construction: the 20/20/10 rule

A first-send sample is always exactly 50 matched NDC records, composed as:

The lower and control records are the point. A file containing only high-variance records demonstrates nothing about method — it is consistent with a correct analysis and equally consistent with selective presentation. Including both directions lets a reviewer check the method rather than take it on faith.

Quality controls

• NDC format validation before any join
• FDA active-registration check on every matched product
• NADAC temporal alignment to the reporting quarter being analyzed
• Outlier boundary validation — implausible unit prices excluded rather than ranked
• Pharmacological cross-reference for NDC reassignment and discontinuation

The computation itself is deterministic SQL against these files. The same inputs and the same query return the same output, which is what allows an agency to reproduce a figure independently.

What this page does not cover

The method above is published so that a reviewing agency can evaluate and reproduce it. The working product is separate and is provided under agreement:

• Ranked candidate files — the full top-250 and top-1,000 outputs for a state or carrier
• Carrier and plan master tables, including parent-organization normalization
• The expansion and prioritization scoring model
• Query logic and the export package
• Multi-state rollups

The distinction is deliberate. A method that cannot be checked is not worth buying, so the method is public. The maintained pipeline, the normalized carrier register, and the ranked output are the engagement.

Limitations

Stating these is part of the method, not a disclaimer bolted to the end of it.

• Managed-care encounter data. SDUD's representation of managed-care encounters is limited, and coverage varies by state and quarter.
• NADAC scope. NADAC surveys retail acquisition cost. Specialty distribution and 340B-acquired product are not represented the same way, and variance involving them should be read with that in mind.
• Fees and contract terms are out of file. Dispensing fees, rebates, and contracted rates are not present in these public sources.
• Vintage sensitivity. A figure is only meaningful alongside the file vintages it came from.
• Coverage. 48 states, as noted above.